1. ISSN: 2278 – 1323
International Journal of Advanced Research in Computer Engineering & Technology (IJARCET)
Volume 2, Issue 4, April 2013
1383
www.ijarcet.org
Abstract— The modern world is enclosed with gigantic
masses of digital visual information. Increase in the images has
urged for the development of robust and efficient object
recognition techniques. Most work reported in the literature
focuses on competent techniques for object recognition and its
applications. A single object can be easily detected in an image.
Multiple objects in an image can be detected by using different
object detectors simultaneously. The paper discusses various
techniques for object recognition and a method for multiple
object detection in an image.
Index Terms— Multi-object detection, Object recognition,
Object recognition applications.
I. INTRODUCTION
The modern world is enclosed with gigantic masses of
digital visual information. To analyze and organize these
devastating ocean of visual information image analysis
techniques are major requisite. In particular useful would be
methods that could automatically analyze the semantic
contents of images or videos. The content of the image
determines the significance in most of the potential uses. One
important aspect of image content is the objects in the image.
So there is a need for object recognition techniques.
Object recognition is an important task in image processing
and computer vision. It is concerned with determining the
identity of an object being observed in an image from a set of
known tags. Humans can recognize any object in the real
world easily without any efforts; on contrary machines by
itself cannot recognize objects. Algorithmic descriptions of
recognition task are implemented on machines; which is an
intricate task. Thus object recognition techniques need to be
developed which are less complex and efficient.
Many successful approaches that address the problem of
general object detection use a representation of the image
objects by a collection of local descriptors of the image
content. Global features provide better recognition. Color and
shape features can also be used. Various object recognition
techniques are presented in this paper. Difficulties may arise
during the process of object recognition; a range of such
difficulties are discussed in this paper. The robust and
efficient object recognition technique can be developed by
Manuscript received Feb, 2013.
Ms.Khushboo Khurana, Computer Science & Engg. department,
S.R.O.C.E.M.., Nagpur, India.
Ms. Reetu Awasthi, Computer Science department, SFS College,
Nagpur, India.
taking into account these difficulties and overcoming them.
Rest of this paper is organized as follows. Section II
elucidates various difficulties in object recognition under
varied circumstances. Section III presents various object
recognition techniques. In Section IV applications for object
recognition are discussed. In section V we have proposed a
method for multi-object detection in an image and finally, we
conclude in Section VI.
II. DIFFICULTIES IN OBJECT RECOGNITION UNDER VARIED
CIRCUMSTANCES
1. Lightning: The lightning conditions may differ during
the course of the day. Also the weather conditions
may affect the lighting in an image. In-door and
outdoor images for same object can have varying
lightning condition. Shadows in the image can affect
the image light. Whatever the lightning may be the
system must be able to recognize the object in any of
the image. Fig.1 shows same object with varying
lightning.
2. Positioning: Position in the image of the object can be
changed. If template matching is used, the system
must handle such images uniformly.
3. Rotation: The image can be in rotated form. The
system must be capable to handle such difficulty. As
shown in fig.2, the character „A‟ can appear in any of
the form. But the orientation of the letter or image
must not affect the recognition of character „A‟ or
any image of object.
4. Mirroring: The mirrored image of any object must be
recognized by the object recognition system.
Khushboo Khurana, Reetu Awasthi
Techniques for Object Recognition in Images
and Multi-Object Detection
Fig.1 Objects with different lightning.
Fig.2 Different orientation of character „A‟

2. ISSN: 2278 – 1323
International Journal of Advanced Research in Computer Engineering & Technology (IJARCET)
Volume 2, Issue 4, April 2013
All Rights Reserved © 2013 IJARCET
1384
5. Occlusion: The condition when object in an image is
not completely visible is referred as occlusion. The
image of car shown in a box in fig.3 is not
completely visible. The system of object recognition
must handle such type of condition and in the output
result it must be recognized as a car.
In [1], a segmentation aware object detection
model is presented with occlusion handling.
6. Scale: Change in the size of the object must not affect
the correctness of the object recognition system.
Above stated are some of the difficulties that may
arise during object recognition. An efficient and robust
object detection system can be developed by conquering
the above stated difficulties.
III. OBJECT RECOGNITION TECHNIQUES
A. Template matching
Template matching is a technique for finding small parts
of an image which match a template image. It is a
straightforward process. In this technique template images
for different objects are stored. When an image is given as
input to the system, it is matched with the stored template
images to determine the object in the input image.
Templates are frequently used for recognition of
characters, numbers, objects, etc. It can be performed on
either color or gray level images.
Template matching can either be pixel to pixel matching
or feature based. In feature based the features of template
image is compared to features of sub-images of the given
input image; to determine if the template object is present
in the input image.
In [2], authors have proposed a mathematical
morphological template matching approach for object
detection in inertial navigation systems (INS). The major
focus of the paper is to detect and track the ground objects.
The flying systems equipped with camera were used to
capture the photos of ground; to identify the objects. Their
method is independent of the altitude and orientation of the
object.
In [3], an approach for measuring similarity between
visual images based on matching internal self-similarities.
A template image is to be compared to another image.
Measuring similarity across images can be complex, the
similarity within each image can be easily revealed with
simple similarity measure, such as SSD (Sum of Square
Differences), resulting in local self-similarity descriptors
which can be matched across images. As shown in fig.4, the
template image of the flower is compared with all the
corresponding descriptors.
B. Color based
Color provides potent information for object
recognition. A simple and efficient object detection scheme
is to represent and match images on the basis of color
histograms.
Fahad Khan, et.al. [4] proposed the use of color
attributes as an explicit color representation for object
detection. The color information is extended in two
existing methods for object detection, the part-based
detection framework and the Efficient Subwindow Search
approach . The three main criteria which should be taken
into account when choosing an approach to integrating
color into object detection are feature Combination,
photometric invariance and compactness. The paper
investigates the incorporation of color for object detection
based on the above mentioned criteria and demonstrate the
advantages of combining color with shape on the two most
popularly used detection frameworks,namely part-based
detection with deformable part models and Efficient
Subwindow Search (ESS) for object localization. The
resulting image representations are compact and
computationally efficient and provide excellent detection
performance on challenging datasets. Fig.5 provide how
the extension correctly detects all Simpsons in the image;
Simpsons is an American animated sitcom. The technique
correctly detects challenging object classes where
state-of-the-art techniques using shape information alone
fail.
Fig.3 Occluded car
(a) (b) (c)
Fig.4. (a) Image template (b) Image against which it is
compared (c) Detected object superimposed on
gray-scale image (from[2])
Fig.5. Find the Simpsons. On the left, the conventional
part based approach fails to detect all four members of
Simpsons. On the right, our extension of the part-based
detection framework with color attributes can correctly
classify all four Simpsons.

3. ISSN: 2278 – 1323
International Journal of Advanced Research in Computer Engineering & Technology (IJARCET)
Volume 2, Issue 4, April 2013
1385
www.ijarcet.org
In [5] the aim is to arrive at recognition of multicolored
objects invariant to a substantial change in viewpoint,
object geometry and illumination. Assuming dichromatic
reßectance and white illumination, it is shown that
normalized color rgb, saturation S and hue H, and the
newly proposed color models c1c2c3 and l1l2l3 are all
invariant to a change in viewing direction, object geometry
and illumination. Further, it is shown that hue H and l1l2l3
are also invariant to highlights. Finally, a change in spectral
power distribution of the illumination is considered to
propose a new color constant color model m1m2m3. To
evaluate the recognition accuracy di¤erentiated for the
various color models, experiments have been carried out on
a database consisting of 500 images taken from 3-D
multicolored man-made objects. The experimental results
show that highest object recognition accuracy is achieved
by l1l2l3 and hue H followed by c1c2c3, normalized color
rgb and m1m2m3 under the constraint of white
illumination. Also, it is demonstrated that recognition
accuracy degrades substantially for all color features other
than m1m2m3 with a change in illumination color. In this
paper the authors aim to examine and evaluate a variety of
color models used for recognition of multicolored objects
according to the following criteria:
1. Robustness to a change in viewing direction
2. Robustness to a change in object geometry
3. Robustness to a change in the direction of the
illumination
4. Robustness to a change in the intensity of the
illumination
5. Robustness to a change in the spectral power
distribution (SPD) of the illumination.
The color models have High discriminative power;
robustness to object occlusion and cluttering; robustness to
noise in the images.
C. Active and Passive
Object detection in passive manner does not involve local
image samples extracted during scanning. Two main
object-detection approaches that employ passive scanning:
1. The window-sliding approach: It uses passive
scanning to check if the object is present or not at all
locations of an evenly spaced grid. This approach
extracts a local sample at each grid point and
classifies it either as an object or as a part of the
background [6].
2. The part-based approach: It uses passive scanning to
determine interest points in an image. This approach
calculates an interest value for local samples at all
points of an evenly spaced grid. At the interest
points, the approach extracts new local samples that
are evaluated as belonging to the object or the
background [7].
Some methods try to bound the region of the image in
which passive scanning is applied. It is a computationally
expensive and inefficient scanning method. In this method
at each sampling point costly feature extraction is
performed, while the probability of detecting an object or
suitable interest point can be squat.
In active scanning local samples are used to guide the
scanning process. At the current scanning position a local
image sample is extracted and mapped to a shifting vector
indicating the next scanning position. The method takes
successive samples towards the expected object location,
while skipping regions unlikely to contain the object. The
goal of active scanning is to save computational effort,
while retaining a good detection performance [8].
The active object-detection method (AOD-method) scans
the image for multiple discrete time steps in order to find an
object. In the AOD-method this process consists of three
phases:
(i) Scanning for likely object locations on a coarse scale
(ii) Refining the scanning position on a fine scale
(iii)Verifying object presence at the last scanning
position with a standard object detector.
D. Shape based
Recently, shape features have been extensively
explored to detect objects in real-world images. The shape
features are more striking as compared to local features like
SIFT because most object categories are better described by
their shape then texture, such as cows, horses and cups and
also for wiry objects like bikes, chair or ladders, local features
unavoidably contain large amount of background mess. Thus
shape features are often used as a replacement or complement
to local features.
A. Berg, et.al. [9], have proposed a new algorithm to find
correspondences between feature points for object
recognition in the framework of deformable shape matching.
The basic subroutine in deformable shape matching takes as
input an image with an unknown object (shape) and compares
it to a model by solving the correspondence problem between
the model and the object. Then it performs aligning
transformation and computes a similarity based on both the
aligning transform and the residual after applying the aligning
transformation. The Authors have considered various reasons
like Intra-category variation, Occlusion and clutter, 3D pose
changes that makes correspondence problems more difficult.
Three kinds of constraints to solve the correspondence
problem between shapes are Corresponding points on the two
shapes should have similar local descriptors, Minimizing
geometric distortion, Smoothness of the transformation from
one shape to the other.
In [10], a new shape-based object detection scheme of
extraction and clustering of edges in images using Gradient
vector Girding (GVG) method is proposed that results a
directed graph of detected edges. The algorithm used contains
a sequential pixel-level scan, and a much smaller second and
third pass on the results to determine the connectiveness. The
graph is built on cell basis and the image is overlaid with a
grid formed of equal sized cells. Multiple graph nodes are
computed for individual cells and then connected
corresponding to the connectivity in the 8-neighbourhood of
each cell. Finally, the maximum curvature of the result paths
is adjusted. The Authors have also proposed several
techniques to increase the performance of the method and

4. ISSN: 2278 – 1323
International Journal of Advanced Research in Computer Engineering & Technology (IJARCET)
Volume 2, Issue 4, April 2013
All Rights Reserved © 2013 IJARCET
1386
improve the quality of the result. The method is particularly
proposed for a RoboCup scenario but can also be applied to
any other images if the prerequisites are met.
In [11] K.Schindler, D. Suter presented a method for
object class detection based on global shape based on elastic
matching of contours. An edge map with closed edge chains is
produced by segmentation into super-pixels. A probabilistic
measure for the similarity between two contours is derived
and then combined with an optimisation scheme to find closed
contours in the image, which have high similarity with a
template image. The method only requires a single object
template. The basic idea used is to define a distance measures
between shapes and then try to find minimum distance. The
Authors discussed different deformable template matching
techniques listed chamfer matching and spline-based shape
matching and compares to elastic matching method and
finally proposed a search strategy which relieves the problem
of false local minima by re-evaluating the shape distance at
each search step.
In [12], boundary structure segmentation (BoSS) is
presented. A ground segmentation method for extractions of
image regions that resembles the global properties of a model
boundary structure and are perceptually prominent. The
global boundary-based shape representation is called
Chordiogram, which is defined as the geometric relationships
between pairs of boundary edges. The perceptual prominence
cue used favour coherent regions different from the
background. The method relates the object detection based on
novel holistic shape descriptor to ground segmentation and
perform them simultaneously. Similarity in shape and
perceptual prominence are the two major properties of
foreground regions selected by BoSS while matching an
image to an object model. The results provide a unified
framework that integrates both object segmentation and
object detection.
E. Local and global features
The most common approach to generic object detection is to
slide a window across the image and to classify each such
local window as containing the target or background. This
approach has been successfully used to detect rigid objects
such as faces and cars in [13] and [6].
In [14], a method of object recognition and segmentation
using Scale-Invariant Feature Transform (SIFT) and Graph
Cuts is presented. SIFT feature is invariant for rotations, scale
changes, and illumination changes. By combing SIFT and
Graph Cuts, the existence of objects is recognized first by
vote processing of SIFT keypoints. Then the object region is
cut out by Graph Cuts using SIFT keypoints as seeds. Both
recognition and segmentation are performed automatically
under cluttered backgrounds including occlusion.
Authors in [15], present a method for object recognition
with full boundary detection by combining affine scale
invariant feature transform (ASIFT) and a region merging
algorithm. The algorithm is invariant to six affine parameters
namely translation (2 parameters), zoom, rotation and two
camera axis orientations. The features give strong keypoints
that can be used for matching between different images of an
object. They trained an object in several images with different
aspects for finding best keypoints of it. Then, a robust region
merging algorithm is used to recognize and detect the object
with full boundary in the other images based on ASIFT
keypoints and a similarity measure for merging regions in the
image. Fig. 6 shows the trained image for an object (left) and
detected image of the object in a image (right).
In [16], Histogram of Gradients (HOG) based multistage
approach for object detection and object pose recognition
for service robots is used. It makes use of the merits of both
multi-class and bi-class HOG-based detectors to form a
three-stage algorithm at low computing cost. In the first
stage, the multi-class classifier with coarse features is
employed to estimate the orientation of a potential target
object in the image; in the second stage, a bi-class detector
corresponding to the detected orientation with intermediate
level features is used to filter out most of false positives; and
in the third stage, a bi-class detector corresponding to the
detected orientation using fine features is used to achieve
accurate detection with low rate of false positives. The
training of multi-class and bi-class support vector machine
(SVM) with their respective features in different levels is
described.
Antonio Monroy, Angela Eigenstetter and Bjorn Ommer
[17], have presented an approach that directly uses
curvature cues in a discriminative way to perform object
recognition. Integrating curvature information substantially
improves detection results over descriptors that solely rely
upon histograms of orientated gradients (HoG). The joint
descriptor is refered as HoGC. Because of the
histogram-nature of the feature vectors, SVM with
histogram intersection kernel is used as a classifier.
An innate extension of these local approaches is to use
sliding window to detect object parts, and then assemble the
parts into a whole object. Problem with local features is that
recognition may fail because of insufficient local
information. This can be solved by using the context of the
image as a whole i.e., global features.
In [18], the gist of an image is computed. First a steerable
pyramid transformation is applied, using 4 orientations and
2 scales; then the image is divided into a 4x4 grid. Object
presence detection determines if one or more instances of
an object class are present. They have combined local
features and global features- GIST for object recognition.
Fig.6. Training image and detected object in the image.
(from [15])

5. ISSN: 2278 – 1323
International Journal of Advanced Research in Computer Engineering & Technology (IJARCET)
Volume 2, Issue 4, April 2013
1387
www.ijarcet.org
IV. APPLICATION OF OBJECT RECOGNITION
1. Biometric recognition: Biometric technology uses
human physical or behavioral traits to recognize any
individual for security and authentication [19].
Biometrics is the identification of an individual
based on distinguished biological features such as
finger prints, hand geometry, retina and iris patterns,
DNA, etc. For biometric analysis, object recognition
techniques such as template matching can be used.
2. Surveillance: Objects can be recognized and tracked
for various video surveillance systems. Object
recognition is required so that the suspected person
or vehicle for example be tracked.
3. Industrial inspection: Parts of machinery can be
recognized using object recognition and can be
monitored for malfunctioning or damage.
4. Content-based image retrieval (CBIR): When the
retrieval is based on the image content it is referred
as CBIR. A supervised learning system, called
OntoPic, which provides an automated keyword
annotation for images and content–based image
retrieval is presented in [20].
5. Robotic: The research of autonomous robots is one of
the most important issues in recent years. The
humanoid robot soccer competition is very popular.
The robot soccer players rely on their vision systems
very heavily when they are in the unpredictable and
dynamic environments. The vision system can help
the robot to collect various environment information
as the terminal data to finish the functions of robot
localization, robot tactic, barrier avoiding, etc. It can
decrease the computing efforts, to recognize the
critical objects in the contest field by object features
which can be obtained easily by object recognition
techniques [21].
6. Medical analysis: Tumour detection in MRI images,
skin cancer detection can be some examples of
medical imaging for object recognition.
7. Optical character/digit/document recognition:
Characters in scanned documents can be recognized
by recognition techniques.
8. Human computer interaction: Human gestures can be
stored in the system, which can be used for
recognition in the real-time environment by
computer to do interaction with humans. The system
can be any application on mobile phone, interactive
games, etc.
9. Intelligent vehicle systems: Intelligent vehicle systems
are needed for traffic sign detection and recognition,
especially for vehicle detection and tracking. In [18],
such a system is developed. In detection phase, a
color-based segmentation method is used to scan the
scene in order to quickly establish regions of interest
(ROI). Sign candidates within ROIs are detected by
a set of Haar wavelet features obtained from
AdaBoost training. Then, the Speeded Up Robust
Features (SURF) is applied for the sign recognition.
SURF finds local invariant features in a candidate
sign and matches these features to the features of
template images that exist in data set. The
recognition is performed by finding out the template
image that gives the maximum number of matches.
V. METHOD FOR MULTI-OBJECT DETECTION IN AN IMAGE
A single image may consist of single or multiple objects. If
all the objects in an image need to be detected the method
shown in fig.7 can be used.
The method trains different object detectors with individual
objects, as shown in fig.7. there are N object detectors which
are trained to detect N different objects. Any of the above
mentioned object recognition techniques can be used
depending upon the application area. An image is provided as
input to the system. The same image is given as input to all
object detectors. Each detector will determine if the object is
present or not. We propose to use object detector along with
boundary detector. If the object is present, the detector will
find its boundary and tag the object name in the image.
So, after the image has passed via all the detectors all objects
will be detected along with object boundary and its tag
displayed in the output image.
Also, when the output image is displayed, we can move the
cursor over the image. The tag shown for an object inside the
complete boundary of the object remains same. Such
multi-object detection in the image can greatly improve the
performance of the content based image retrieval systems.
The performance can further be improved by letting the
object detectors run in parallel.
VI. CONCLUSION
In this paper, we have discussed various object detection
techniques. The template matching technique requires large
database of image templates for correct object recognition.
Hence it must be used only when limited objects are to be
detected. Global features and shape based method can give
Fig.7. Method for multi-object detection in an image

6. ISSN: 2278 – 1323
International Journal of Advanced Research in Computer Engineering & Technology (IJARCET)
Volume 2, Issue 4, April 2013
All Rights Reserved © 2013 IJARCET
1388
better result and are efficient as compared to local features.
These techniques help in easy access of the images. They also
find their application in fields such as biometric recognition,
medical analysis, surveillance, etc. A method for multiple
object detection is also presented.
REFERENCES
[1] T. Gao, B. Packer, D. Koller, “A Segmentation-aware Object Detection
Model with Occlusion Handling”, In IEEE International Conference on
Computer Vision and Pattern Recognition (CVPR), June 2011.
[2] W. Hu, A.M.Gharuib, A.Hafez, “Template Match Object Detection for
Inertial Navigation Systems,” Scientific research (SCIRP), pp.78-83,
May 2011.
[3] E.Shectman, M.Irani, “Matching Local Self-Similarities across Images
and Videos,” In IEEE International Conference on Computer Vision
and Pattern Recognition, pp. 1-8, 2007.
[4] F. Khan, R. Muhammad , et.al., “Color Attributes for Object
Detection,” In IEEE International Conference on Computer Vision and
Pattern Recognition (CVPR), pp. 3306 – 3313, 2012.
[5] T. Gevers, A. Smeulders, “Color-based object recognition,” Pattern
Recognition, 1999.
[6] P. Viola and M. Jones, “Robust real-time object detection,”
International. Journal of Computer Vision, 57(2), pp.137–154, 2004.
[7] R. Fergus, P. Perona, A. Zisserman, “Weakly supervised
scale-invariant learning of models for visual recognition,”
International Journal of Computer Vision, 2006.
[8] G. de Croon, “Active Object Detection,” In 2nd International
conference on computer vision theory and applications (VISAPP
2007), Barcelona, Institute for Systems and Technologies of
Information, Control and Communication (INSTICC), pp. 97–103,
2007.
[9] A. Berg, T.Berg , J. Malik, “Shape Matching and Object Recognition
using Low Distortion Correspondences,” In IEEE International
Conference on Computer Vision and Pattern Recognition (CVPR), pp.
26 – 33, 2005.
[10] H. Moballegh, N. Schmude, and R. Rojas, “Gradient Vector Griding:
An Approach to Shape-based Object Detection in RoboCup
Scenarios,” from:
www.ais.uni-bonn.de/robocup.de/papers/RS11_Moballegh.pdf
[11] K.Schindler, D. Suter, “Object Detection by Global Countour Shape,”
Pattern Recognition, 41(12), pp.3736–3748, 2008.
[12] F. Khan, R. Muhammad , et.al., “Color Attributes for Object
Detection,” In IEEE International Conference on Computer Vision and
Pattern Recognition, pp. 3306 – 3313, 2012.
[13] C. Papageorgiou and T. Poggio, “A trainable system for object
detection,” International. Journal of Computer Vision, 38(1),
pp.15–33, 2000.
[14] A. Suga, K. Fukuda, T. Takiguchi, Y.Ariki, “Object Recognition and
Segmentation Using SIFT and Graph Cuts,” In 19th International
Conference on Pattern Recognition ,pp. 1-4, 2008 .
[15] R. Oji, “An Automatic Algorithm for Object Recognition and
Detection Based on ASIFT Keypoints,” Signal & Image Processing:
An International Journal (SIPIJ) Vol.3, No.5, pp.29-39, October 2012.
[16] L. Dong, X. Yu ,L. Li, J. Kah Eng Hoe, “HOG based multi-stage object
detection and pose recognition for service robot,” Control Automation
Robotics & Vision (ICARCV), 11th International Conference, pp.
2495 – 2500, Dec. 2010.
[17] A. Monroy, A. Eigenstetter, B. Ommer, “Beyond Straight Lines -
Object Detection Using Curvature,” In 18th IEEE International
conference on Image Processing (ICIP), pp. 3561 - 3564 , 2011.
[18] K. Murphy, A. Torralba , D. Eaton and W. Freeman, “Object detection
and localization using local and global features,” Towards
Category-Level Object Recognition, 2005.
[19] V. Bjorn, “One Finger at a Time: Best Practices for Biometric
Security,” Banking Information Source (Document ID: 1697301411),
April, 2009.
[20] J. Schober, T. Hermes, O. Herzog, “Content-based Image Retrieval by
Ontology-based Object Recognition,” In KI Workshop on
Applications of Description Logics , 2004.
[21] W. Chang, C. Hsia. Y. Tai, et.al, “An efficient object recognition
system for humanoid robot vision,” Pervasive Computing (JCPC),
IEEE, December, 2009.
Khushboo Khurana received her B.E. degree in
computer science and engineering from RTM Nagpur
University, Nagpur, in 2010. She is currently pursuing
her M.Tech. (CSE) from Shri Ramdeobaba College of
Engineering and Management (Autonomous), Nagpur.
Her interests include image and video processing.
Reetu Awasthi received her M.Sc. (Computer
Science) degree from RTM Nagpur University,
Nagpur, in 2012. She is currently working as a lecturer
in SFS College, Nagpur. Her interests include
biometric security and image processing.